flow

/flow-decompose-project

Phase: Any (standalone)

Usage:

/flow:flow-decompose-project <project description>
/flow:flow-decompose-project --step N --id <id>

Decomposes a large project into many GitHub issues with native sub-issue relationships, blocked-by dependencies, milestones, and phase labels. Produces a fully linked issue graph ready for autonomous execution via /flow-start or /flow-orchestrate.


What It Does

Each step is enforced via self-invocation — the skill re-invokes itself with --step N --id <id> after each gate, forcing the model to re-read the full skill instructions at every step boundary. The <id> is a short UUID generated in Step 1 that scopes all file paths to prevent concurrent session collisions.

Step Name Gate
1 Describe and Decompose AskUserQuestion: proceed, iterate, or cancel
2 Review Issue List AskUserQuestion: due date + approve, revise, or cancel
3 Create Epic and Milestone Automatic
4 Create Child Issues Automatic
5 Link Relationships Automatic
6 Report Summary + cleanup
  1. Step 1 — Describe and Decompose: Invokes decompose:decompose with deep codebase exploration. Presents the DAG synthesis for user review.
  2. Step 2 — Review Issue List: Builds a complete issue list from the DAG with titles, bodies, phase labels, and dependencies. Asks for milestone due date. User approves the full list before any issues are created.
  3. Step 3 — Create Epic and Milestone: Creates the milestone with the due date and the parent epic issue.
  4. Step 4 — Create Child Issues: Creates all child issues in topological order (leaves first) so dependency numbers exist when referenced. Each issue gets the “Decomposed” label and an auto-derived phase label.
  5. Step 5 — Link Relationships: Sets sub-issue relationships (children to epic) and blocked-by dependencies (between children per DAG) via GitHub REST API. Best-effort throughout.
  6. Step 6 — Report: Presents a summary table of everything created, then cleans up session files.

Issue Format

Each child issue contains:


GitHub Relationships

The skill creates three types of GitHub relationships:

All relationship creation is best-effort. Native blocked-by relationships are the sole dependency mechanism — analyze-issues queries them via GraphQL to detect blocked issues.


Gates